After a wafer is scanned by a charged particle beam, there are numerous scanned images respectively corresponds to numerous chips of the wafer. Clearly, some scammed images are defect-free images when the corresponding chips have no defect, and other scanned images are defective images when each corresponds chip has at least one defect.
There are some known arts to find which scanned image is defective image and to find which portion(s) of the defective image corresponds to the defect(s). For example, die-to-die, die-to-database, and array mode. However, almost all these known arts achieve the results by comparing the scanned images corresponding to the same chip. Hence, these known arts only can find which portion(s) of which chip corresponds to defect(s), but can not find what kind of the defect(s) is (at least can not find what semiconductor structure corresponds to the defect(s)). In short, it cannot identify a defect from a lead, a MOS or an electrode under the situation of lacking the layout of the chip. Even the defect feature is important for improving the manufacturing process.